Szilard Pall

Szilárd Páll is an HPC software scientist at the KTH Center for Scientific Computing (KCSC). He works on parallel algorithms and software for high-performance computational science. With a background in computer science and computational biophysics, he has worked on GPU acceleration of scientific codes since 2008.

He is the principal investigator of the SSF-funded SEESSI project (2026–2029) and is building a research team that develops algorithms for data movement, communication, and reduced and mixed precision for the flagship codes GROMACS, Neko and VeloxChem, with energy to solution as a primary metric.

As a core developer of GROMACS, he helped reformulate key molecular-dynamics algorithms for modern processor architectures and was an early architect of its heterogeneous CPU-GPU parallelization. He continues to work on its parallelization design, most recently GPU-resident simulation and GPU-initiated communication for strong scaling. He has also helped drive its GPU portability work, from the original OpenCL backend to the current SYCL backend.

He brings emerging hardware and programming models into production scientific software in close collaboration with hardware vendors. This includes a long-standing collaboration with NVIDIA on GROMACS, which he leads. He has also applied his parallel algorithm work beyond molecular dynamics, contributing to the heterogeneous parallelization design and scalability of the Neko CFD code (ACM Gordon Bell Prize finalist, SC23).

His current work focuses on GPU-native communication for strong scaling of molecular-dynamics simulations, performance portability and mixed-precision algorithms. He co-organizes the APerOL minisymposium series on performance-portable scientific software.